Split losthdr between client and backend

[Lasse Karstensen]

Hello.

I'd like to propose an additional varnishstat counter that counts requests rejected
(413) due to http_req_hdr_len overflow.

When we do Varnish diagnostics for customers that are having problems, we often see
increased losthdr. It would be very helpful to know if it was a broken client
(usually loads of cookies) or a backend response that made this counter
increase.

In the case of a client you can increase http_req_hdr_(len|size), and in the backend
case you probably want to fix your backend.

Knowing which solution to recommend would be a big plus.

I had a look at the code and adding this seems pretty simple to put into
cache/cache_http1_fsm.c's http1_dissect(). (trivial patch here: http://pastie.org/5661064)

However, I'm not convinced that this is the best place to do this, or that this
is the best way of solving my initial problem. I'd appreciate any input on this.

-- 
With regards,
Lasse Karstensen
Varnish Software AS